Herd behavior is a phenomenon that often appears in the stock market.It is caused by the irrational imitation of investors and is expressed as major investors make similar investing decisions in a short time period.Detecting herd behavior is necessary for both regulatory authorities and investors to reduce investment risk.Previously, many researchers focused on herd behavior detection and proposed numbers of detection methods.Although most of these methods only can judge whether herd behavior exists in the stock market, they cannot detect which stocks are in the same herd and which stock is the leader of the herd.In this paper, we proposed an efficient herd behavior detection approach based on density clustering.It firstly uses stock return rates in a period to calculate similarity of stock-pairs at that time.Then, it finds the leader of herds by finding local density peaks.After that, it clusters stocks following their probable leaders and discovers herds in stock markets.Finally, it determines whether existing herd behavior and the degree of herd behavior according to the size of herds, i.e.the number of stock in herd.Experimental results on the Chinese stock market demonstrate the effectiveness and accuracy of this approach.